Recent observations of the forest treeline and highest altitude individual trees in the Swiss Alps indicate a consistent and accelerating upward shift over the past several decades. These shifts can result in changes to the composition and structure of ecosystems in the Alps. As new tree species establish at higher elevations, they may outcompete or replace native tree and plant species. This can potentially lead to disruptions in the ecological balance and resilience of Alpine forests as well as key functions such as carbon sequestration and nutrient cycling.
While warming trends are known to affect the limits and spatial distribution of Alpine vegetation colonies, the full scope of the mechanisms driving colonisation and growth patterns of tree species at upper elevation limits are not well understood.
Can you summarise the theme and objective of your project?
Jesse Ray Murray Lahaye (J.L.): To better understand the mechanisms driving colonisation and growth patterns, ecological modeling frameworks must be developed to encompass the governing processes behind these drivers. However, a significant challenge arises when attempting to scale such models to regional levels due to the substantial amount of detailed observations needed for their validation and calibration. In many cases, acquiring such data using ground or satellite-based techniques is not feasible or possible due to cost, accessibility, or limited data resolution.
In this work, we approached this challenge by employing airborne remote sensing data to create a comprehensive high-resolution tree species inventory dataset covering an entire mountain valley in Western Switzerland. This was achieved through the implementation of a data-driven learning approach.
How is it innovative in your field of research and has a significant impact in the context of sustainability and climate action?
J.L.: By combining airborne imagery, laser scanning data, and in-situ ground truth observations, we developed a deep learning-based image processing method capable of automatically generating an accurate and high-resolution forest inventory dataset that describes the entire vertical transect of our study area. Currently ongoing work to utilise this data to calibrate and validate a set of process-based Species Distribution Models (SDMs) developed to predict the future evolution of the Alpine treeline ecotone, with the goal of improving their robustness and scalability is currently ongoing the results of which will represent the first time models of these types have been refined with such high resolution and broad scaled data.
In a few words, what are the results or applications of your project?
J.L.: The image processing framework developed during this project facilitated the automatic detection and characterisation of nearly 50,000 trees, achieving an accuracy of over 70% in determining their crown width and height, in line with ground based measurement techniques. This is in comparison to a four month long, in-situ ground truthing campaign that was conducted over the course of the project where only approximately 900 trees were inventoried, albeit with a significant amount of effort and time. This generated inventory will be subsequently used in the calibration and validation of SDMs designed to forecast the future evolution of the Alpine treeline ecotone at regional scales in our study area. The predictions generated through this work will serve as valuable information to local and regional forest and environmental management agencies in their efforts to maintain the ecosystem functions within this region in response to climate change.
In addition to the forest and environmental management agencies, to what public/industry/society stakeholders will your results be directly useful? What impact do you hope to have in the medium and long term?
J.L.: The final results of our study will be communicated through the EPFL internal departmental communication services and have the potential for publication in at least two journals in domains ranging from ecology to geodesy and remote sensing. The environmental monitoring capacity and data that will be developed within this project will be released as open data and will thus serve as a valuable resource for on-going joint initiatives of EPFL and UNIL (CLIMACT MOUNTEGAL, BlueMount initiatives), as well as to future work with collaboratives such as the EPFL Center for Imaging.
The findings of our study will be disseminated to the general public through publication in national and local newspapers (e.g. Le Temps, Heidi.news, NZZ, Le Nouvelliste) and through outreach events organised through Valais based public education centers such as the Musée de la Nature in Sion (e.g. during the « Nuit des Musées ») and the alpine garden Flore-Alpe in Champex. Outreach activities will be coordinated with Alpole, CLIMACT and Mediacom on the EPFL side and with the UNIL Interdisciplinary Centre for Mountain Research and the Service for Culture and Outreach. Exhibitions highlighting the findings of our project hosted at Flore-Alpe, EPFL and UNIL have the potential to move within Switzerland (visitor centers of regional or national parks) and within the Alps (e.g. Museum of Grenoble).
Could you tell us how you collaborated with researchers from UNIL and EPFL on your project?
J.L.: The collaborative effort between researchers from the Department of Ecology and Evolution (DEE) at UNIL, the Geodetic Engineering Laboratory (TOPO) and the Environmental Computational Science and Earth Observation Laboratory (ECEO) at EPFL played a pivotal role in our project, demonstrating the importance of interdisciplinarity in addressing complex ecological questions. The TOPO laboratory contributed its expertise in the acquisition and processing of spatially referenced data, which was essential for producing the Analysis Ready Data (ARD) required for our project. In parallel, the ECEO laboratory provided their extensive proficiency in applying deep learning methods to environmental and remote sensing data.
In what way was interdisciplinarity essential?
J.L.: This expertise was crucial in developing a scalable model that effectively linked airborne data to species-level colonisation patterns. Working in tandem, the DEE at UNIL brought their specialised knowledge in modeling the factors that explain spatio-temporal species distribution evolution, and integrating the remote sensing based forest inventory toward the enhancement of the SDM frameworks.
The distinct yet complementary areas of expertise from our research groups allowed us to explore and test hypotheses that were previously unresolved in the field of evolutionary ecology. By leveraging novel Machine Learning (ML) techniques and Earth observation technologies, we created potential to uncover new insights and potential solutions to intricate ecological challenges.
Team behind the research:
Jesse Ray Murray Lahaye: Doctoral assistant at the Laboratory of Cryospheric Sciences (CRYOS)
Prof. Devis Tuia: Associate Professor, Environmental Computational Science and Earth Observation Laboratory (ECEO)
Prof. Jan Skaloud: Senior scientist at ENAC's Environmental Science and Engineering Section and at the Doctoral Program in Civil and Environmental Engineering (EDCE) EDCE, and Host Professor at the Laboratory of Cryospheric Sciences (CRYOS)
Dr. Christophe Randin: Researcher at the Department of Ecology and Evolution (DEE) of UNIL, on the climate impact on mountain trees' spring phenology, the exploration of rare alpine plant survival under climate shifts, and the analysis of plant dynamics in urbanised Canton de Vaud using historical specimens
Find out more about CLIMACT: